Quick glossary
VU (Virtual User)
Virtual user simulated in parallel by k6/JMeter/Gatling. Load metric #1.
Ramp-up
Progressive climb from 0 → target load. Gives autoscalers time to react.
Plateau
Load held at target. Where you measure stability and SLOs.
P95 (latency)
95% of requests answer faster than this. Better than average for UX.
P99 (latency)
Catches the slowest requests (cold-starts, GC pauses, locks). Reveals anomalies.
RPM / RPS
Requests per minute (or per second). Effective throughput absorbed by the system.
Throughput
Volume of requests handled over time. Measure on plateau, not peak.
SLO
Service Level Objective. Numeric target to honor (e.g. P95 ≤ 500ms, errors ≤ 1%).
项目上下文
上下文越精准，建议越锋利
Everything flows through a single deployable. Typical bottlenecks: shared DB, locks, transactions too wide.
压测参数
设定目标负载与持续时间
用户场景
百分比合计需为 100%（当前：100%）
总动作量估计
~52 500
在 20 分钟内（ramp-up + plateau）
目标负载
500
吞吐
3 000/min
总时长
20 分钟
百分比一致性
100%
合计 100%
按场景拆分
在 Timint App 中升级到专业版
分析 P95/P99、错误率、得分；获取专家修复建议；生成精美 PDF 审计报告。
✅ 自动计算性能得分
✅ 性能与云架构专家建议
✅ 带图表的 PDF 报告 + 个人主页保存
Detailed shot plan
Total and per-scenario load at each ramp-up step, then on plateau
| Phase | Timing | Total VU | Total req/min | search-cart (75%) | checkout (20%) | signup (5%) |
|---|---|---|---|---|---|---|
| Ramp-up step 1/4 (25%) | T+1.3 分钟 | 125 | 750 | 94 ·563 | 25 ·150 | 6 ·38 |
| Ramp-up step 2/4 (50%) | T+2.5 分钟 | 250 | 1 500 | 188 ·1 125 | 50 ·300 | 13 ·75 |
| Ramp-up step 3/4 (75%) | T+3.8 分钟 | 375 | 2 250 | 281 ·1 688 | 75 ·450 | 19 ·113 |
| Ramp-up step 4/4 (100%) | T+5.0 分钟 | 500 | 3 000 | 375 ·2 250 | 100 ·600 | 25 ·150 |
| Sustained plateau | T+5→20 分钟 | 500 | 3 000 | 375 ·2 250 | 100 ·600 | 25 ·150 |
| Estimated volume: ~7 500 actions during ramp-up + ~45 000 actions during plateau = ~52 500 requests total. | ||||||
Risks per critical service
4 service(s) checked — each exposed service has its own risk profile
Risk认证
Auth saturation = session blocks and refresh-token loops. Watch: users-table DB pool, JWT cache, /login throttling.
Risk搜索
Search latency → instant cart abandonment. Bottlenecks: Elasticsearch / Algolia indexes, facet queries, full-text scans.
Risk结算
Checkout failure = direct revenue loss. Measure the full funnel (cart → shipping → payment → confirmation).
Risk支付
PSP error = lost payments or double charges. Idempotency keys are mandatory, PSP A → B failover in < 60s.
Expert recommendations — 单体
DevOps, SRE and Solution Architect tips tailored to your architecture
🛠️ DevOps & Cloud scale
- •Profile the 10 most expensive endpoints before the test (APM: Datadog, NewRelic, Sentry Performance)
- •Enable DB slow query log and archive it during the test for post-mortem analysis
- •Provision +30% CPU/RAM during the test if stateful — otherwise scale horizontally with round-robin LB
- •Freeze deployments and disable cron jobs during the test window
⚙️ SRE / Platform
- •Define the 4 golden signals BEFORE the test: latency, traffic, errors, saturation
- •Alert on SLOs (P95, error rate), not raw CPU thresholds
- •Prepare a read-only degraded DB mode in case of unexpected saturation
- •Test rollback of the previous deployment in under 5 minutes
🏛️ Solution architect
- •Identify bounded contexts (DDD) to prepare a potential split into modular monolith
- •Progressively move heavy jobs (PDFs, emails, exports) to a queue + worker
- •Externalize sessions to Redis before scaling horizontally (stateless first)
- •Measure operational cost per feature to target the next split
Expert advice tailored to the project — 电商
Recommendations specific to the business constraints of your project type
Ready-to-copy k6 code
Script auto-generated with your parameters + scenarios + SLO thresholds
import http from 'k6/http';
import { sleep, group } from 'k6';
export const options = {
stages: [
{ duration: '1.3m', target: 125 },
{ duration: '1.3m', target: 250 },
{ duration: '1.3m', target: 375 },
{ duration: '1.3m', target: 500 },
{ duration: '15m', target: 500 },
],
thresholds: {
http_req_duration: ['p(95)<500', 'p(99)<1500'],
http_req_failed: ['rate<0.01'],
},
};
export default function () {
group('search-cart', () => {
// 75% du trafic — 375 VU cible · 2250 req/min
// TODO: implement requests for the search-cart scenario
});
group('checkout', () => {
// 20% du trafic — 100 VU cible · 600 req/min
// TODO: implement requests for the checkout scenario
});
group('signup', () => {
// 5% du trafic — 25 VU cible · 150 req/min
// TODO: implement requests for the signup scenario
});
sleep(10.0);
}Pre-shot checklist
Tick before launching to avoid false alarms
Share this simulation
Generate a public link to send to your team, client or board
The link will be public. Only the values entered here will be published. No personal data beyond the optional name.
一次规划周密的 load test 比一次生产故障便宜 10 倍。在这里规划，用你的工具（k6、JMeter、Locust）执行，再到 Timint App 里分析并生成可分享的审计报告。
Timint Smart Tips
AI勾选了支付服务：务必使用 PSP 沙箱，并准备 3DS retries / idempotency keys。
电商 + 搜索：把热门 facets 建索引，热门查询加短缓存。
下载 Timint App 来分析结果并生成 PDF 审计报告。
Timint App: 强大精确的模拟工具，助您做出最佳财务决策。计算、打印、保存 — 完全免费。
FAQ
更进一步
Timint 应用内 35+ 模拟器
在 Timint 应用中找到您的所有工具、实时收入计数器以及更多功能。支持 iOS 和 Android。